ee9d9 - 4h

GA Nostr. Thought of the day: In praise of "repetition" in learning. Sorry for polluting your feed with LinkedIn content, but Michele is good people (and I’m doing whatever I can to get him to try better social media). https://www.linkedin.com/posts/michelesollecito_doing-the-same-exercise-1000-times-is-way-activity-7343907900572004352-9XR2 I’m 100% with him here. Step 1: Pick a problem. Step 2: Work on it. (Slowly… What’s the rush? This is learning on your own time). Step 3: Go back to Step 2. Get stuck. Unlock yourself. Get deep into it. Repeat. Repeat. Repeat. Repeat. Repeat. Get stuck again. Get bored. Repeat. Give up. Start over with the "perfect architecture" after all the lessons learned... Turns out it wasn’t that perfect after all. Get stuck. Unlock yourself. Undo a bunch of things. Save what you can. Throw away the rest. Challenge your assumptions. Figure out where the real value is. Understand not just the tech stack, but the actual problem you’re trying to solve. Actually learn something! It turns out that once you truly understand what’s involved in building something, you’ll realise that the five-minute, vibe-coded app (even your classic to-do list, blog engine, or snake game) is just a toy—and it stays a toy even after months or years of work. Learn how to do it anyway. Learn how to build it, run it, maintain it. Learn how to deal with users, bugs, feature requests, security issues, outages, costs, community, contributions (if OSS), and so on. In a time of increasing immediatism, fake urgency, and widespread Dunning–Kruger effect, keep your ego in check and take time to learn for real. You won’t just be one of the few people who can actually build something in tech... As others fall into lazy wishful thinking and helplessness, you’ll be among the small percentage who can actually do things. There’s no shortcut. There’s no end. It’s a struggle. And it’s immensely fun. Just do it. #GM #SoftwareEngineering #Learning #RealTalk #NoShortcuts #MasteryTakesTime

0
1
3

b2b13 - 2y

0
0
0

d6858 - 2y

0
0
0

0d6f3 - 2y

因为从短期来说,解决眼前的矛盾和困惑 就是生物最优先的问题。所以所有的感受和情绪才会寻求最直接的解决方式,短期内 最直接的解 基本都会很接近最优解,特别是对生存有影响的问题。 而长期的多因素的 对错(是否有更优解)的问题,是后来人类生命周期延长和社会化后才需要面对的问题。 至于AI训练 ,也没超出过这个框架,只是底层的硬件还远未到可以做更复杂的实时决策(或近实时,不然用个brute force跑到天荒地老也可以称为AGI)。现在的AI其实就是一堆记忆能力超强的神经元,所形成的系统复杂度远在昆虫以下。但不代表这些记忆力强的神经元 被运用得当的话,实际的能力会很弱。可以把现在的chatgpt当作一种更高级的程序模块,然后可以通过finetune+数据库把模块专向化,最后人为地把这些模块组装成一个更庞大的系统。(虽然难很多很多倍,但操作逻辑上和现在的软件工程没什么区别)。 展望一下,我觉得这样做出来的东西能达到 “99.99% 的人在聊1小时内 无法分辨是不是bot” "有初中以上的综合学历"。这种东西虽然做不了科研,发掘不了什么新知识,但对绝大部分人而言已经足够AGI了。 可以参考以下这个很粗糙的wolfram整合,还有现在bing的AI chatbot扒资料整理的能力。这两者都只是很简易地把不同的程序整合在一起,所以我认为以后系统级的 整合 和finetune后,潜力巨大。 https://writings.stephenwolfram.com/2023/01/wolframalpha-as-the-way-to-bring-computational-knowledge-superpowers-to-chatgpt/ #AI #ChatGPT #DeepLearning #SoftwareEngineering #Biology #Intelligence #SoftwareIntegration

0
0
0

Showing page 1 of 1 pages